// 处理对应的dbBatch
private DbBatch getDbBatch(HttpPipeKey key) {
String dataUrl = key.getUrl();
Pipeline pipeline = configClientService.findPipeline(key.getIdentity().getPipelineId());
DataRetriever dataRetriever =
dataRetrieverFactory.createRetriever(
pipeline.getParameters().getRetriever(), dataUrl, downloadDir);
File archiveFile = null;
try {
dataRetriever.connect();
dataRetriever.doRetrieve();
archiveFile = dataRetriever.getDataAsFile();
} catch (Exception e) {
dataRetriever.abort();
throw new PipeException("download_error", e);
} finally {
dataRetriever.disconnect();
}
// 处理下有加密的数据
if (StringUtils.isNotEmpty(key.getKey()) && StringUtils.isNotEmpty(key.getCrc())) {
decodeFile(archiveFile, key.getKey(), key.getCrc());
}
InputStream input = null;
JSONReader reader = null;
try {
input = new BufferedInputStream(new FileInputStream(archiveFile));
DbBatch dbBatch = new DbBatch();
byte[] lengthBytes = new byte[4];
input.read(lengthBytes);
int length = ByteUtils.bytes2int(lengthBytes);
BatchProto.RowBatch rowbatchProto =
BatchProto.RowBatch.parseFrom(new LimitedInputStream(input, length));
// 构造原始的model对象
RowBatch rowBatch = new RowBatch();
rowBatch.setIdentity(build(rowbatchProto.getIdentity()));
for (BatchProto.RowData rowDataProto : rowbatchProto.getRowsList()) {
EventData eventData = new EventData();
eventData.setPairId(rowDataProto.getPairId());
eventData.setTableId(rowDataProto.getTableId());
eventData.setTableName(rowDataProto.getTableName());
eventData.setSchemaName(rowDataProto.getSchemaName());
eventData.setEventType(EventType.valuesOf(rowDataProto.getEventType()));
eventData.setExecuteTime(rowDataProto.getExecuteTime());
// add by ljh at 2012-10-31
if (StringUtils.isNotEmpty(rowDataProto.getSyncMode())) {
eventData.setSyncMode(SyncMode.valuesOf(rowDataProto.getSyncMode()));
}
if (StringUtils.isNotEmpty(rowDataProto.getSyncConsistency())) {
eventData.setSyncConsistency(SyncConsistency.valuesOf(rowDataProto.getSyncConsistency()));
}
// 处理主键
List<EventColumn> keys = new ArrayList<EventColumn>();
for (BatchProto.Column columnProto : rowDataProto.getKeysList()) {
keys.add(buildColumn(columnProto));
}
eventData.setKeys(keys);
// 处理old主键
if (CollectionUtils.isEmpty(rowDataProto.getOldKeysList()) == false) {
List<EventColumn> oldKeys = new ArrayList<EventColumn>();
for (BatchProto.Column columnProto : rowDataProto.getOldKeysList()) {
oldKeys.add(buildColumn(columnProto));
}
eventData.setOldKeys(oldKeys);
}
// 处理具体的column value
List<EventColumn> columns = new ArrayList<EventColumn>();
for (BatchProto.Column columnProto : rowDataProto.getColumnsList()) {
columns.add(buildColumn(columnProto));
}
eventData.setColumns(columns);
eventData.setRemedy(rowDataProto.getRemedy());
eventData.setSize(rowDataProto.getSize());
// 添加到总记录
rowBatch.merge(eventData);
}
dbBatch.setRowBatch(rowBatch);
input.read(lengthBytes);
length = ByteUtils.bytes2int(lengthBytes);
BatchProto.FileBatch filebatchProto =
BatchProto.FileBatch.parseFrom(new LimitedInputStream(input, length));
// 构造原始的model对象
FileBatch fileBatch = new FileBatch();
fileBatch.setIdentity(build(filebatchProto.getIdentity()));
for (BatchProto.FileData fileDataProto : filebatchProto.getFilesList()) {
FileData fileData = new FileData();
fileData.setPairId(fileDataProto.getPairId());
fileData.setTableId(fileDataProto.getTableId());
fileData.setEventType(EventType.valuesOf(fileDataProto.getEventType()));
fileData.setLastModifiedTime(fileDataProto.getLastModifiedTime());
fileData.setNameSpace(fileDataProto.getNamespace());
fileData.setPath(fileDataProto.getPath());
fileData.setSize(fileDataProto.getSize());
// 添加到filebatch中
fileBatch.getFiles().add(fileData);
}
dbBatch.setFileBatch(fileBatch);
return dbBatch;
} catch (IOException e) {
throw new PipeException("deserial_error", e);
} finally {
IOUtils.closeQuietly(reader);
}
}
// ======================== help method ===================
// 保存对应的dbBatch
private HttpPipeKey saveDbBatch(DbBatch dbBatch) {
RowBatch rowBatch = dbBatch.getRowBatch();
// 转化为proto对象
BatchProto.RowBatch.Builder rowBatchBuilder = BatchProto.RowBatch.newBuilder();
rowBatchBuilder.setIdentity(build(rowBatch.getIdentity()));
// 处理具体的字段rowData
for (EventData eventData : rowBatch.getDatas()) {
BatchProto.RowData.Builder rowDataBuilder = BatchProto.RowData.newBuilder();
rowDataBuilder.setPairId(eventData.getPairId());
rowDataBuilder.setTableId(eventData.getTableId());
if (eventData.getSchemaName() != null) {
rowDataBuilder.setSchemaName(eventData.getSchemaName());
}
rowDataBuilder.setTableName(eventData.getTableName());
rowDataBuilder.setEventType(eventData.getEventType().getValue());
rowDataBuilder.setExecuteTime(eventData.getExecuteTime());
// add by ljh at 2012-10-31
if (eventData.getSyncMode() != null) {
rowDataBuilder.setSyncMode(eventData.getSyncMode().getValue());
}
if (eventData.getSyncConsistency() != null) {
rowDataBuilder.setSyncConsistency(eventData.getSyncConsistency().getValue());
}
// 构造key column
for (EventColumn keyColumn : eventData.getKeys()) {
rowDataBuilder.addKeys(buildColumn(keyColumn));
}
// 构造old key column
if (CollectionUtils.isEmpty(eventData.getOldKeys()) == false) {
for (EventColumn keyColumn : eventData.getOldKeys()) {
rowDataBuilder.addOldKeys(buildColumn(keyColumn));
}
}
// 构造其他 column
for (EventColumn column : eventData.getColumns()) {
rowDataBuilder.addColumns(buildColumn(column));
}
rowDataBuilder.setRemedy(eventData.isRemedy());
rowDataBuilder.setSize(eventData.getSize());
rowBatchBuilder.addRows(rowDataBuilder.build()); // 添加一条rowData记录
}
// 处理下FileBatch
FileBatch fileBatch = dbBatch.getFileBatch();
BatchProto.FileBatch.Builder fileBatchBuilder = null;
fileBatchBuilder = BatchProto.FileBatch.newBuilder();
fileBatchBuilder.setIdentity(build(fileBatch.getIdentity()));
// 构造对应的proto对象
for (FileData fileData : fileBatch.getFiles()) {
BatchProto.FileData.Builder fileDataBuilder = BatchProto.FileData.newBuilder();
fileDataBuilder.setPairId(fileData.getPairId());
fileDataBuilder.setTableId(fileData.getTableId());
if (fileData.getNameSpace() != null) {
fileDataBuilder.setNamespace(fileData.getNameSpace());
}
if (fileData.getPath() != null) {
fileDataBuilder.setPath(fileData.getPath());
}
fileDataBuilder.setEventType(fileData.getEventType().getValue());
fileDataBuilder.setSize(fileData.getSize());
fileDataBuilder.setLastModifiedTime(fileData.getLastModifiedTime());
fileBatchBuilder.addFiles(fileDataBuilder.build()); // 添加一条fileData记录
}
// 处理构造对应的文件url
String filename =
buildFileName(rowBatch.getIdentity(), ClassUtils.getShortClassName(dbBatch.getClass()));
// 写入数据
File file = new File(htdocsDir, filename);
OutputStream output = null;
try {
output = new BufferedOutputStream(new FileOutputStream(file));
com.alibaba.otter.node.etl.model.protobuf.BatchProto.RowBatch rowBatchProto =
rowBatchBuilder.build();
output.write(ByteUtils.int2bytes(rowBatchProto.getSerializedSize())); // 输出大小
rowBatchProto.writeTo(output); // 输出row batch
com.alibaba.otter.node.etl.model.protobuf.BatchProto.FileBatch fileBatchProto =
fileBatchBuilder.build();
output.write(ByteUtils.int2bytes(fileBatchProto.getSerializedSize())); // 输出大小
fileBatchProto.writeTo(output); // 输出file batch
output.flush();
} catch (IOException e) {
throw new PipeException("write_byte_error", e);
} finally {
IOUtils.closeQuietly(output);
}
HttpPipeKey key = new HttpPipeKey();
key.setUrl(remoteUrlBuilder.getUrl(rowBatch.getIdentity().getPipelineId(), filename));
key.setDataType(PipeDataType.DB_BATCH);
key.setIdentity(rowBatch.getIdentity());
Pipeline pipeline = configClientService.findPipeline(rowBatch.getIdentity().getPipelineId());
if (pipeline.getParameters().getUseFileEncrypt()) {
// 加密处理
EncryptedData encryptedData = encryptFile(file);
key.setKey(encryptedData.getKey());
key.setCrc(encryptedData.getCrc());
}
return key;
}
@Override
public void extract(DbBatch dbBatch) throws ExtractException {
Assert.notNull(dbBatch);
Assert.notNull(dbBatch.getRowBatch());
// 读取配置
Pipeline pipeline = getPipeline(dbBatch.getRowBatch().getIdentity().getPipelineId());
boolean mustDb = pipeline.getParameters().getSyncConsistency().isMedia();
boolean isRow = pipeline.getParameters().getSyncMode().isRow(); // 如果是行记录是必须进行数据库反查
// 读取一次配置
adjustPoolSize(pipeline.getParameters().getExtractPoolSize()); // 调整下线程池，Extractor会被池化处理
ExecutorCompletionService completionService = new ExecutorCompletionService(executor);
// 进行并发提交
ExtractException exception = null;
// 每个表进行处理
List<DataItem> items = new ArrayList<DataItem>();
List<Future> futures = new ArrayList<Future>();
List<EventData> eventDatas = dbBatch.getRowBatch().getDatas();
for (EventData eventData : eventDatas) {
if (eventData.getEventType().isDdl()) {
continue;
}
DataItem item = new DataItem(eventData);
// 针对row模式，需要去检查一下当前是否已经包含row记录的所有字段，如果发现字段不足，则执行一次数据库查询
boolean flag =
mustDb
|| (eventData.getSyncConsistency() != null
&& eventData.getSyncConsistency().isMedia());
// 增加一种case, 针对oracle erosa有时侯结果记录只有主键，没有变更字段，需要做一次反查
if (!flag && CollectionUtils.isEmpty(eventData.getUpdatedColumns())) {
DataMedia dataMedia = ConfigHelper.findDataMedia(pipeline, eventData.getTableId());
if (dataMedia.getSource().getType().isOracle()) {
flag |= true;
eventData.setRemedy(true); // 针对这类数据，也统一视为补救的操作，可能erosa解析时反查数据库也不存在记录
}
}
if (isRow && !flag) {
// 提前判断一次，避免进入多线程进行竞争
// 针对view视图的情况，会有后续再判断一次
flag = checkNeedDbForRowMode(pipeline, eventData);
}
if (flag
&& (eventData.getEventType().isInsert()
|| eventData.getEventType().isUpdate())) { // 判断是否需要反查
Future future =
completionService.submit(new DatabaseExtractWorker(pipeline, item), null); // 提交进行并行查询
if (future.isDone()) {
// 立即判断一次，因为使用了CallerRun可能当场跑出结果，针对有异常时快速响应，而不是等跑完所有的才抛异常
try {
future.get();
} catch (InterruptedException e) {
cancel(futures); // 取消完之后立马退出
throw new ExtractException(e);
} catch (ExecutionException e) {
cancel(futures); // 取消完之后立马退出
throw new ExtractException(e);
}
}
futures.add(future); // 记录一下添加的任务
}
items.add(item); // 按顺序添加
}
// 开始处理结果
int index = 0;
while (index < futures.size()) { // 循环处理发出去的所有任务
try {
Future future = completionService.take(); // 它也可能被打断
future.get();
} catch (InterruptedException e) {
exception = new ExtractException(e);
break; // 如何一个future出现了异常，就退出
} catch (ExecutionException e) {
exception = new ExtractException(e);
break; // 如何一个future出现了异常，就退出
}
index++;
}
if (index < futures.size()) {
// 小于代表有错误，需要对未完成的记录进行cancel操作，对已完成的结果进行收集，做重复录入过滤记录
cancel(futures);
throw exception;
} else {
// 全部成功分支, 构造返回结果也要保证原始的顺序
for (int i = 0; i < items.size(); i++) {
DataItem item = items.get(i);
if (item.filter) { // 忽略需要被过滤的数据，比如数据库反查时记录已经不存在
eventDatas.remove(item.getEventData());
}
}
}
}
